Source: Geek Park

Author: Jesse

Editor: Jingyu

Image source: Generated by Wujie AI

The wheel of productizing large models is rolling faster and faster.

Recently, after quickly completing the cold start with AI-generated portraits, Miaoya has launched a new feature: AI photo retouching. The new feature allows what used to take an hour for a girl to retouch her beauty in photos to be completed in seconds, and it can even solve problems such as closed eyes, messy hair, and facial expressions.

On the surface, retouching may not seem as magical and topical as "portrait generation." However, in reality, retouching is a more high-frequency product point with a potentially broader market. Over the past decade, various retouching and beauty software have proven to be a mature use case with solid demand.

The launch of AI retouching by Miaoya signifies its transformation from a technically demo-oriented experimental project to an internet product oriented towards demand. Compared to traditional retouching and filter tools, AI intervention may completely revolutionize the experience of such products. This will inevitably lead to a fierce collision between Miaoya and retouching apps including Meitu Xiuxiu, just like the war between short video live streaming and e-commerce entering each other's territories in the past.

This collision is not only a change in the vertical field of beauty retouching but also reflects the future prospect of "AI challenging the old world." So, what fundamental changes can AI bring to the art of retouching?

01 A "High-Frequency Hard Scenario"

In July of this year, "Miaoya Camera" was quietly launched without prior warm-up or marketing.

Relying on the "auto-generated portraits" feature, Miaoya completed the cold start within a few weeks, mostly relying on user-initiated discussions and sharing on social media. Because users tend to share the generated portraits on platforms like WeChat Moments after uploading their own photos.

Miaoya became the first product in the domestic AIGC field to "break the circle."

Looking back at Miaoya's early success, on the one hand, it was due to its technological foundation. Although the team currently keeps their model implementation method strictly confidential, the fact that there hasn't been a product on the market that can compete with it after six months of Miaoya's launch is enough to demonstrate its technological leadership.

On the other hand, Miaoya found a sufficiently viral scenario and quickly broke the barriers to dissemination through product design.

However, it must be acknowledged that "portrait generation" is actually a low-frequency scenario. After trying it out a few times, users did not have the demand or motivation to continue using Miaoya.

Many early users would say that the generated portraits by Miaoya could already rival those produced by studios like Haimat and Tianzhenlan. However, whether it's a formal portrait or an ID photo, their use cases mostly lie in the public social sphere, making it difficult for AI-generated photos to replace real portraits.

Therefore, for Miaoya, they had almost no time to celebrate the influx of the first wave of traffic. In order to retain users and generate subsequent growth through dissemination, product innovation must not stop for a moment.

Finally, Miaoya chose "retouching" as a new entry point.

This is not difficult to understand. Firstly, "retouching" is a concept that has been developed for many years, and "optimizing image effects with algorithms" has been integrated into the entire production process of photos, from computational photography on smartphones, beauty camera apps, to post-editing retouching and filter software…

Secondly, after years of development, retouching has proven to be a very high-frequency hard demand, and this field has given rise to listed companies like Meitu, and even Instagram, whose initial product was a "retro filter camera."

Lastly, the retouching field has already established a mature market and business model, and many users are motivated to pay. This is crucial for Miaoya, an AI company with high model training costs, as they have determined from the beginning that "AIGC products must start charging from day one."

02 Winning with "Speed"

For Miaoya, the launch of the "retouching" feature is a huge challenge.

This is a very mature product track that has undergone more than a decade of full development and competition, with every point along the entire pipeline already filled with products.

From the computational photography algorithms of smartphone manufacturers, the beauty retouching functions of original cameras, to various third-party cameras, filters, beauty apps, and even the image editing tools built into social media… It can be said that this is a market with a low entry barrier and oversaturation. Users have too many choices, and most of them are free.

Through AI large models, Miaoya has adopted a completely new technological foundation, but how to translate this technology into a perceptible product advantage for users is still a complex problem.

Miaoya decided to win with "speed."

A typical user's photo-taking scenario today is as follows: after finding a good scene and posing, everyone will take many photos in succession, then switch poses and continue taking more photos. Among the tens or hundreds of photos, finding one with the right angle, reasonable lighting, appropriate composition, and natural expression, and then retouching the skin, facial features, and hairstyle.

For many girls, the process of retouching photos may involve first selecting a few from many, applying basic filters, and doing rough retouching; then selecting one from many, and retouching them one by one. The entire process often takes several tens of minutes to start.

For AI retouching, its biggest advantage is being light, fast, and burden-free, from uploading photos to completing retouching, everything can be solved with just one click.

On social networks, people have already begun fervently discussing this magical product. Some users have posted comparison photos of closed eyes and the beautiful photos after AI retouching, exclaiming, "My discarded photos have been saved"; while others have posted a comparison photo of messy hair that was fixed with just one click by AI, expressing amazement, "What technology and skill is this," "Will retouchers lose their jobs?"

For users, this means that the threshold and startup cost of putting a photo into the "retouching" process have been lowered, allowing them to take more photos, send more photos to retouching apps, and ultimately "produce" faster.

Just as the comments from users suggest, AI retouching also solves the "scale" problem of traditional retouching.

Traditional retouching can adjust facial features, skin flaws, but it is basically unable to handle problems like closed eyes, uncontrolled expressions, etc. This is why many people take many photos, first selecting a few with good expressions, and then proceeding to retouch them.

For AI retouching, whether it's messy hair, covering the face, or facial expression distortion, all can be solved by AI's computing power.

In Miaoya's product, in addition to the options of "light retouch" and "beautification," there are also options for "refreshing" and "reshaping," corresponding to situations of facial expression distortion and even complete obstruction of the face by hair, which can be repaired by AI. This allows many "discarded photos" to become useful, and also eliminates the need to take many photos and make selections during the shooting process.

Based on AIGC technology, not only has the retouching process for users been simplified, but it can even save users time and energy during the shooting process.

03 Continuously Evolving in Competition

After the launch of the retouching feature, Miaoya clearly set Meitu Xiuxiu as its benchmark, indicating that the two will enter into competition. Meitu Xiuxiu also recently launched a module similar to Miaoya's AI-generated portraits, and the intention of both sides to enter each other's "territory" is very obvious.

As the leader in the field of retouching apps, Meitu has been in business for 15 years. According to the latest financial report data, Meitu has 240 million monthly active users, 5.6 million paying users, and total revenue exceeding 2 billion yuan in 2022. On the other hand, Miaoya is a project that has just been born less than half a year ago and is still in its early stages.

But perhaps this is the embodiment of the "acceleration" of large-scale model entrepreneurship. It is driven by the exponential expansion of parameters and model algorithm capabilities.

In fact, AI retouching is not exclusive to Miaoya, as more and more companies are targeting image applications based on AIGC. Even Meitu itself has been exploring AIGC since last year and recently launched an AI portrait feature similar to Miaoya's camera, which is basically the same in terms of product logic.

At the same time, Meitu is also preparing to launch new AIGC features such as "AI enlargement" and "AI animation" in the second half of this year.

With the launch of AI retouching by Miaoya, the two companies are inevitably going to face each other head-on, initiating a "total war."

Just like the competition between new energy vehicles and traditional fuel vehicles that is currently in full swing, the ultimate result, whether for the former or the latter, is aimed at the direction of "intelligence" evolution.

But this is the inevitable process in the development of technology. When technology enters a new development period and leaps to a new level, everyone will arrive at a new starting line.

Now, the tide of large models and AIGC has already arrived, and every company must take the plunge.

The key in the end still lies in how to cultivate the technological foundation, find the best product entry point, and enter a healthy, constantly evolving development track.

This is also the biggest challenge that Miaoya will face in the future.

04 The Greater Value of "AI Retouching"

After Miaoya launched the "AI portrait" feature, some controversial voices emerged: Are the AI-generated photos still the real you? Many users have expressed their astonishment, saying, "The generated photos look too good, to the point where I can't believe it's me."

In fact, this contradiction has existed since the first day of the birth of "retouching software." Users record their "copies" with a camera, and what they want is ultimately "truth" or "beauty"? The greatest value of AI retouching may lie here—telling users from the algorithm's perspective what the "golden balance point between truth and beauty" is.

Today, many beauty effects are becoming more and more exaggerated, almost evolving into a "currency inflation of beauty tools."

This has led to many conflicts. One view is that beauty retouching is like makeup, a reasonable pursuit of beauty by people; while another view believes that beauty retouching makes people increasingly afraid to face their true selves, and begins to criticize the practice of excessive photo retouching.

The contradiction between these two voices has even gradually evolved into a social issue. Today, the time spent on retouching photos after taking them is getting longer, often in search of a balance point that "looks beautiful but still maintains a natural truth."

And this contradiction, perhaps, can be reconciled through AI.

Although the "digital doppelganger" generated by AI makes people look beautiful, don't forget that it is still based on the user's uploaded real photos.

We can actually expect AI tools to establish a baseline of "at least looking like yourself after retouching." Although this baseline is still relatively vague at the moment, with the improvement of data and parameters, it will definitely become clearer and more accurate, and the results of retouching will become more and more "like yourself."

In fact, the Miaoya team has already noticed the early users' astonishment at the generated portraits being "too good to believe it's themselves," and has made improvements to address this issue in the retouching feature. The current AI retouching feature includes a "light retouch" option, which is a relatively restrained level. The team's focus on model optimization in the future will be on grasping the scale of beauty retouching.

After more than a decade of the birth of beauty retouching functions, many people would say that contemporary people have lost "diverse beauty." For example, people with freckles and moles on their faces only have the option to "smooth them out" in beauty camera apps, and the only beautification direction for the skin is whitening and brightening.

In fact, many celebrities and models, even with freckles and not fair skin, can turn those flaws into a unique expression of beauty when captured by excellent, professional photographers, paired with appropriate clothing and accessories, those flaws can also become a unique expression of beauty, and people with dark skin can also highlight their contours…

It can be imagined that based on the computing power of AI large models, future retouching software may be able to help us explore "diverse beauty."

With the improvement of parameters and the accumulation of data, AI retouching may become a "new mirror" for us, helping us discover a new self, not just limited to skin smoothing, face slimming, whitening, and enlarging the eyes…

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。